2023-07-04    Share on: Twitter | Facebook | HackerNews | Reddit

Easy Text Vectorization With VectorHub and Sentence Transformers

Learn how to use Sentence Transformers for text vectorization with different models using consistent API.

Text is heavily inspired by part of the e-book: Semantic NLP search with FAISS and VectorHub - Guide To Vectors (getvectorai.com) - which was using VectorHub as an interface to the models.

NOTE: VectorHub is deprecated and no longer maintained. The authors of VectorHub recommend using Sentence Transformers, TFHub, and Huggingface directly for text vectorization.

This article demonstrates a similar process as the original article but uses a sentence transformers package.

Encoding Data Using Sentence Transformers

To encode models easily, we will utilize the Sentence Transformers library. SentenceTransformers is a Python framework for state-of-the-art sentence, text, and image embeddings. It provides a variety of pre-trained models that can convert sentences into meaningful numerical representations.

First, we need to install the sentence-transformers package, which includes the necessary dependencies for using Sentence Transformers. This library offers a wide range of pre-trained models, such as BERT, RoBERTa, and MiniLM, that can be used for text encoding. More information about Sentence Transformers can be found here.

pip install sentence-transformers

Next, we will instantiate our model and start the encoding process. In this example, we will use the "all-MiniLM-L6-v2" model, which is a variant of the MiniLM model.

from sentence_transformers import SentenceTransformer

model = SentenceTransformer('all-MiniLM-L6-v2')

# Sentences to be encoded
sentences = [
    'This framework generates embeddings for each input sentence',
    'Sentences are passed as a list of strings.',
    'The quick brown fox jumps over the lazy dog.'

# Encode sentences using the Sentence Transformers model
embeddings = model.encode(sentences)

# Print the embeddings
for sentence, embedding in zip(sentences, embeddings):
    print("Sentence:", sentence)
    print("Embedding:", embedding)

In the code snippet above, we begin by installing the sentence-transformers package, which provides the necessary tools for working with Sentence Transformers. This library offers various pre-trained models that can convert sentences into meaningful vector representations.

After the installation, we import the SentenceTransformer class from the sentence_transformers module. We instantiate the model using the all-MiniLM-L6-v2 variant, which will be used for encoding our sentences.

We define a list of sentences that we want to encode using the Sentence Transformers model. In this case, we have three exemplary sentences: "This framework generates embeddings for each input sentence," "Sentences are passed as a list of strings," and "The quick brown fox jumps over the lazy dog."

To perform the encoding, we use the encode method of the model object, passing in the sentences list. This method returns the corresponding embeddings for each sentence, which we store in the embeddings variable.

Finally, we iterate over the sentences and embeddings lists together using zip. For each sentence and its corresponding embedding, we print them out to visualize the results.

Please note that the code snippet above uses the "all-MiniLM-L6-v2" model as an example. You can explore and choose from a wide range of models provided by Sentence Transformers according to your specific requirements.


Any comments or suggestions? Let me know.

To cite this article:

    author  = {Krystian Safjan},
    title   = {Easy Text Vectorization With VectorHub and Sentence Transformers},
    journal = {Krystian's Safjan Blog},
    year    = {2023},